POWERED
A predictive health and maintenance app
for critical energy equipment
Probabilistic Operations Warranted for Energy
Reliability Evaluation and Diagnostics (POWERED)
The performance of energy equipment degrades over time. A power supply failure in a critical situation can severely impact residents, commercial businesses, and local, state, and federal Government. POWERED is a prognostic and diagnostic app for energy equipment that can help operators maintain power supplies and protect against future risks.
POWERED uses rich, modular probabilistic modeling to help operators determine the reliability of power supplies before a crisis occurs. POWERED integrates large amounts of available transformer data to diagnose equipment health and predict the possibility of system failures. Such proactive maintenance can prevent catastrophic shutdowns and save millions of dollars lost to expensive downtime and damaged equipment.
Dr. Sanja Cvijic
Senior Scientist and Principal Investigator on POWERED
AI-based predictive maintenance
Digital instrumentation on today’s transformers generate thousands of daily datapoints that can be used for AI–based predictive analytics. Dissolved gas analysis, a technique used to diagnose transformer health, used to be done every six to twelve months. Today’s tools can conduct oil sampling every hour.
POWERED analyzes a variety of data: ambient temperature, electrical loading, temperature of the oil used as cooling encasing agent, composition of dissolved gases in oil, and more.
Physics-based or data-driven models alone prove ineffective in capturing the diversity of variables to be analyzed. POWERED’s hybrid AI approach extends both types of models as needed: physics models that capture physical phenomena and data-driven simulations when physics models are not available. Instead of developing a model from scratch, our tool reuses different models and chains them together to run end-to-end inference on the unified model. To facilitate this integration, the tool uses Scruff™, Charles River Analytics’ probabilistic and modeling framework.
The dashboards receive processed data from the Central Server after predictive analytics have been performed
on the latest supervisory control and data acquisition (SCADA) measurements.
Powering energy-informed operations
for reliable and efficient supply
POWERED’s results are easy to visualize and facilitate faster decision-making. The dashboard is modular and provides full visibility into what the model believes at any time—explainable AI in action. Users are able to mix and match variables to study their correlations at any point in time. The “Monitoring” dashboard shows historical measurements as solid lines and predicted quantities as dashed lines, while the “What-If” dashboard allows the user to modify predicted quantities—such as loading, ambient temperature, and through faults—and helps analyze their system-wide effects.
POWERED is driven by the Army’s overall vision of energy-informed operations that provide a reliable and uninterrupted power supply with improved operational efficiency. Through our initial effort, we proudly partnered with City Light & Power, Inc., provider of electric utility systems. Through our ongoing development, we have extended the tool to ingest power quality data, visualize that data, and provide advanced data analytics.
While POWERED was originally created as a tool for the military, distribution and power companies represent a significant group of potential beneficiaries. After iterative user-tested improvements, POWERED has the potential to expand beyond servicing transformers and can be geared toward different types of equipment.
Contact us to learn more about POWERED and our capabilities in probabilistic programming and energy efficiency.
This material is based upon work supported by the Small Business Innovative Research (SBIR) Program and the Engineering Research and Development Center (ERDC) – Construction Engineering Research Laboratory (CERL) under Contract W9132T-17-C-0018. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the SBIR Program and ERDC-CERL.